Lars Eggert
Lars Eggert
Making this a draft until the test failure is fixed.
Removing the early return makes it not pass the `amplificationlimit` test anymore though. Sigh.
This again is waiting for a fix to the `idle_timeout_crazy_rtt` test failure.
> With this change, the server can only update the RTT to a small value: > > ``` > 0s202ms INFO [pri-unv-path:0800000000b3b9a6 [fe80::1]:443->[fe80::1]:443] discarding a packet without an RTT estimate;...
Some QNS L1/C1 tests are failing because #1981 is needed.
So this seems to work, in the sense that there are no more `A` failures when we are the server. But maybe there are also more `L1` and `C1` failures...
I have another reproducer. In #1875, the `idle_timeout_crazy_rtt` fails with this every time at the moment.
The issue in `idle_timeout_crazy_rtt` in #1875 seems to be that the idle timeout is the same as `now`: ``` 0s 26ms TRACE [Server a437d00156e81143] Get callback delay Instant { tv_sec:...
@stefson is this still an issue?
RACK and/or other CC improvements are attractive areas of work, but I think we first need to be sure that the underlying machinery (timers, buffers, loss recovery logic, etc.) is...